Published August 28, 2026 | Version v1.0

TheDongLab/AI2AMP-PD: AI2AMP-PD v1.0

  • 1. Aligning Science Across Parkinson's (ASAP) Collaborative Research Network, Chevy Chase, MD, USA
  • 2. Stephen and Denise Adams Center for Parkinson's Disease Research of Yale School of Medicine, New Haven, CT 06511, USA

Description

This is the public release of the analysis code accompanying the manuscript "Differential expression and machine-learning classification of Parkinson's disease using AMP-PD RNA-seq data" (Hu R, Dong X, et al.).

Overview

Analysis pipeline for RNA-seq–based differential expression (DE) and machine-learning (ML) classification of Parkinson's disease (PD) case/control status, using AMP-PD cohorts PPMI and PDBP/BioFIND (PDBF).

What's included

  • Data preparation — build case/control and mutation-carrier sample sets and extract per-cohort RNA-seq quantification matrices (DE/src/filtration.py, DE/src/extract_quant.py)
  • Differential expression — DESeq2-based DE analysis for genes, eRNAs, and circRNAs, adjusted for covariates (age, sex, plate, RIN, genotype PCs) (DE/src/DE_PCs.R, DE/src/eRNA/, DE/src/circRNA/)
  • Functional enrichment — GO/KEGG/Reactome/GSEA analysis on DE results (DE/src/Enrich_Profiler.R)
  • ML classification — logistic regression, LASSO, SVM, random forest, XGBoost, KNN, MLP, and stepwise feature-addition classifiers trained on PPMI and independently tested on PDBF (ML/src/*.py)
  • Model evaluation — combined AUROC/PR curve and feature-importance summary plots (ML/src/31_plots_AUC_PR_inOne.py)
  • NanoString QC — standalone endogenous/housekeeping gene QC script (DE/NanoStringData/)
  • Recorded R/Python session info and pinned dependencies (session_info/)

Requirements

  • R ≥ 4.0 (tidyverse, DESeq2, clusterProfiler, and related Bioconductor/CRAN packages)
  • Python ≥ 3.8 (pandas, numpy, scikit-learn, matplotlib, seaborn)

Known limitations

  • Scripts are tailored to the specific AMP-PD sample sets and covariate schemas used in this study; not general-purpose tools
  • File paths and cohort-specific parameters are set via literals/CLI args and may need editing for new datasets
  • Raw AMP-PD data are not distributed with this repository; access requires a separate application to the AMP-PD Knowledge Portal (https://amp-pd.org/)

License

MIT License

Files

TheDongLab/AI2AMP-PD-v1.0.zip

Files (168.0 MB)

Name Size
md5:24441e665fdb6d03be3cc47bec79fe5b
168.0 MB Preview Download

Additional details

Related works

Funding

Aligning Science Across Parkinson's
Parkinson5D: deconstructing proximal disease mechanisms across cells, space, and progression ASAP-000301